[KLUG Hardware] Re: hardware RAID cards (was Informal Survey)

[Mike Williams]

Bryan J. Smith wrote:

>On Wed, 2002-10-23 at 15:45, Mike Williams wrote:  
>  
>
>>SRAM is faster than SDRAM I take it?  I'm a bit fuzzy on the
>>difference other than SRAM doesn't lose everything when unpowered.
>>    
>>
>
>Not so much "faster" as in "throughput" but "faster" as in "latency." 
>It's the reason why 100MHz SRAM L2 Cache on a board is faster than
>100MHz SDRAM Memory is on the same board.
>
>DRAM memory takes, typically, 40-70ns to access, which is a 4-7 clock
>cycle delay than sub-10ns SRAM, which is 0 cycle delay.  After a few
>reads, those delays add up big time.
>  
>
At 4-7 clock cycles per read, ouch.

>>I don't suppose you have a link to any articles at StorageReview? 
>>    
>>
>
>Their review database has been trashed a few times.
>  
>
A storage website with storage problems of their own?  Doesn't exactly 
inspire confidence.

>>Well, at least you diminished mine a bit from giving me that article
>>to proofread.  [incoming rant] I find it a bit scary that most of the
>>high-end system vendors these days (Alienware, Falcon NW, Voodoo, and
>>anybody else who advertises in MaxPC) build BIOS RAID 0 into most of
>>their machines.  They reviewed one a couple months ago that had FOUR
>>120 gig drives in a stripe set!  Passing on the question of  why
>>anybody would need 1/2 a terabyte in a workstation / game system, it
>>seemed like a dumb idea to me to let any of your 4 drives take down
>>the whole array like that.  For an almost $8000 system (a lot of that
>>being 24 inch LCD) wouldn't another $200 for reduntant 360 gigs be
>>better than a house-of-cards 480?  [end rant]Which nicely brings me to
>>the main question I was hoping you could answer.  Since all out speed
>>was the goal on this machine, that's the only reason I can see for not
>>giving it RAID 5 with that many drives.  So any idea how much speed do
>>you actually lose going from RAID 0 to RAID 5?
>>    
>>
>
>Depends on the number of writes.  If the writes are fairly sequential, a
>3Ware Escalade card with 1-4MB of SRAM can easily handle RAID-5, and you
>only lose 1 disk in efficiency, plus the small XOR overhead.  If you
>have lots of random writes, then it really hurts.
>
How about reads, since a workstation does more read than write?  In 
sequential reads I could see the sorting of access requests that SCSI 
allows (forgot the technical term) being an advantage.  Is the 
controller on an Escalade smart enough to do that sort of thing with IDE?

>Actually, I sure wish 3Ware would do RAID-4, it would be better for lots
>of large, sequential writes.  I've suggested it to them before.
>
My aforementioned ignorance is showing again.  I looked up levels 3 and 
4 and couldn't tell the difference.  They both seem to be RAID 5, but 
with all parity data on the same drive.  What's the difference between 3 
and 4?  The advantage to it being that reads don't have to hit the 
parity drive at all, and sequential writes keep the parity data 
contiguous, right?